Now, let's compare machine learning in Finance with machine learning in Tech in terms of tasks. First, let's start with applications of machine learning in Tech. Most common applications of machine learning in Tech are image recognition and various natural language processing tasks such as text classification, sentiment analysis, machine translation and so on. According to our specification of machine learning and to perception tasks, and action tasks, this applications belong in the class of perception tasks. This is because for these tasks there is only a fixed single action that the algorithm has to perform well. For example to translate the text from one language to another, this problem are usually solved using the supervise learning, sometimes in the combination with unsupervised learning. This, for example so fetch in tasks in Tech. We can mention various problems in computational advertising, robotics, and self driving cars. For this task, the reinforcement learning provides the most natural framework and is used for methods in practice, along with methods based on supervised and unsupervised learning. Now let's see what kind of perception and action tasks are encountered in Finance. First, there are many types of forecasting problems in Finance that belong in the class of perception tasks. They include, in particular, predictions of prices of securities, such as stocks, bonds, or commodities. For this problem, supervised and unsupervised learning are the methods of choice. Other forecasting tasks in Finance include predicting actions of individual players. For example, we may want to predict actions of corporate, such as dividend payments, defaults, mergers and so on. Another problem is prediction of actions of individuals with ample defaults and loans, fraud, or anti-money-laundering. It turns out that for such prediction tasks, we can use not only supervised and unsupervised learning, but also reinforcement learning. Other typical perception tasks which are commonly used in Finance deal with valuation. For example, valuation of assets such as stocks, futures, commodities, and bonds is related to prediction problems, but is different from it. Another classic valuation problem in Finance is derivative pricing. Even though the most conventional way for derivatives pricing is via using parameterized stochastic models, the problem lends itself quite naturally for machine learning. In the last course of this specialization, we will see how reinforcement learning can be used to price financial derivatives. Now let's note something interesting here. Remember that in Tech, perception tasks are solved using supervised and unsupervised learning, while reinforcement learning is only used for action tasks. Yet, in Finance, it sounds as even perception tasks can employ methods based on reinforcement learning. We can ask, what is the source of such difference between perception tasks in Finance and Tech? And the short answer I can offer would be this. In Finance, perception tasks are often about the future. For example, about prediction of dividends paid by company in the next quarter. However, this future is driven, to a large extent, by specific decisions of decision makers. For example, next quarter dividends are not some sort of random stochastic process, but rather decided by a company management. And this is exactly the reason reinforcement learning might be a good conceptual framework in this subject. This is because reinforcement learning directly models the process of decision-making. Therefore, in the example of corporate dividends, it tries to explain the physics of the process rather blindly, if you like, from a purely statistical approach. Now, let's consider action tasks in Finance. Typical example of this in Finance would be in trading and asset management. Including the problems of optimal execution for brokerage trading, optimal strategies for day trading, or active portfolio management. Because they hold you with the problem of choosing optimal action, reinforcement learning is a very natural framework for these problems. Further examples of action task in Finance involves various decision making processes in banking. For example loan approvals, credit and operational risk management or decision making in compliance analytics, or anti-money laundering would all be examples of action tasks. Again, reinforcement learning is a very naturall framework for such settings. So let's summarize what we found so far. We found perception tasks in Finance might sometimes be very different from perception tasks in Tech. Such difference arises for forecasting tasks that involve predicting future actions, so in rational or semi-rational market agents. For example, actions of corporations or individuals. Reinforcement network is a framework that fits very naturally in such problems. Moreover the fact that the reinforcement learning appears as a method of choice for both perception and action tasks in Finance unlike Tech. When it only appears in action tasks, point to important qualitative differences between machine learning in Finance and machine learning in Tech. In addition, there are also some important quantitative differences that we will discuss next.
